SSP Ensemble Review

Seasonal Patterns in Sea Surface Temperatures

Author

Adam Kemberling

Published

August 31, 2023

Reviewing the Bias-Corrected Ensemble Datatsets from CMIP6

This report exists to visually review the bias-corrected CMIP6 data ensembles. Ensembles are from two Shared Socio-economic Pathways (SSPs): 1. SSP 1 (Sustainability) 2.6
2. SSP 5 (Fossil Fueled Development) 8.5

This document will look into three things. 1. How do these two pathways compare for our study areas 2. How do they compare against real-life observations 3. Are there regional biases or errors

Additionally, I also plan to flag points in time when each SSP ensemble passes temperature thresholds for the different areas. Bias corrected timelines for the sea surface temperature, sea surface salinity, bottom temperature, and bottom salinity will be compared against the reference observations (OISSTv2, SODA) to check how accurately they represent observed conditions, and whether that changes between our areas of interest.

Focus Areas

For clarity on what data is included/attributed to each region, the areas we focused on have been plotted below.

Loading the Ensemble and Historical Datasets

Bias-Corrected SSP Ensemble Data

Both the bias-corrections and the observational datasets (OISSTv2, SODA) have been prepared separately and are loaded below.

CMIP6 Bias-Corrected Arrays

Monthly Averaging
Remove Individual Model Runs

In this step we are streamlining the data to just the ensemble mean/5th/95th percentiles instead of keeping the individual model runs along for the ride.

Bias-Corrected Ensemble -

Bias Correction Observation Period = (2000-2019)

The following figures chart out how each variable changes in time across the different models within an area. This data has been locally bias-corrected using observational datasets (OISSTv2) and reanalysis models (SODA) using data over an observation period of 2000-2019 which is based on data availability limitations (no bottom temperature before 2000).

Historic Observation Datasets

There are two data sources with which the CMIP6 model outputs are checked against to perform the bias-corrections. These are the OISSTv2 and SODA reanalysis datasets. Monthly-averaged timeseries for each region can be viewed below.

Comparing Scenarios to Observations

One of the goals of performing bias-correction is to address any systematic differences (biases) that can be seen between an ensemble scenario’s data and real-life data. If an ensemble appears to be “running hot” (always hotter than real-life) then we can bring those values down and hopefully have better expectations about future conditions.

The following figures look at how real life observation data (black lines) fall between the upper and lower expected ranges for the two scenarios.

Enhanced Biological Data

To fit the VAST model a suite of covariates matching the biological observations are uploaded. These values should fall within the ranges of the observed Temp/Salinity values and SSP ranges.

This section looks at the location-specific observations that are collected as part of the biological surveys. For all of the specific locations & times where biological sampling occurs we record what the surface/bottom conditions are at those sites.

Because survey sampling locations are selected at random, these values may be higher/lower than what the average across what the broader area may be. However, these values are expected to be in the same ballpark. Large differences here may be indicative of sampling biases towards areas with specific conditions that aren’t represented well by the broader regional average. OR they might indicate bugs in the data processing that extracts environmental conditions for these points.

In the following figures the points are the within-year averages of the point-extracted seasonal covariates. Point-locations come from actual survey locations. Data is extracted for each point for the relevant seasonal

How do the un-scaled seasonal values look?

Before going into the VAST model, all the inputs are in the units that they are measured originally (temperature, ppt). They should all be directly comparable and consistent within regions here.

VAST covariate scaling and prediction un-scaling:

Before we fit the model, we center and scale each variable. At this point the trends should resemble the above figures, but the y-axis values should now be in scaled units closer to 0. These are the units that the VAST model looks at when looking at relationships between the presence and abundance of a species and changes in the physical environment.

Rescaled Enhanced Biological Data and Projected Physical Data –

This dataset is at > Box/Mills Lab/Projects/COCA19_Projections/data/combined/all_tows_all_covs_rescale.rds.”

We keep the mean/SD info for each variable and then also go through a process to rescale the projected values from the CMIP6 experiments using these mean and SD values before making the projections.

These “prediction” datasets are located in > Box/Mills Lab/Projects/COCA19_Projections/data/predict”

They are named: 1. CMIP6_SSP5_85_mean_dat.csv 2. CMIP6_SSP5_85_95thpercentile_dat.csv 3. CMIP6_SSP5_85_5thpercentile_dat.csv 4. CMIP6_SSP1_26_mean_dat.csv 5. CMIP6_SSP1_26_95thpercentile_dat.csv 6. CMIP6_SSP1_26_5thpercentile_dat.csv

How do the Scaled Seasonal Values look?

Distance from Modern Observations: 2010-2019

A 20-year period was used as a “baseline period” for fitting species distribution models to environmental features of surface and bottom temperatures in space. Each ensemble scenario was bias-corrected using real-life observational data to align scenario data for this period with the range of temperatures/salinities that species are experiencing currently.

As a way to assess how far from a contemporary baseline we’ve moved, we can compare future conditions against a modern baseline of 2010-2019. In the following tables the average annual temperatures for this period have been removed, with temperatures reported as a departure from this baseline.

These tables could be slimmed down a lot. I think the most interesting things to include would be the means from both scenarios, and the distance between the two of them: “SSP1: 2.6 projects a mean temperature of +1C by 2040 in the Gulf of Maine. The more pessimistic scenario SSP5: 8.5 projects a similar future at +1.2C”… Something like that. Range of confidence and amount they overlap would also be interesting

Ensemble Differences from Modern Conditions
Difference in Bias-Corrected Ensemble Data when Compared to 2010-2019 Averages from OISSTv2 & SODA
SSP1: 2.6 SSP5: 8.5
Ensemble Mean 5th Percentile 95th Percentile Ensemble Mean 5th Percentile 95th Percentile
Canadian Survey Area - Surface Temperature
2000 −0.94 1.03 1.03 −0.94 1.03 1.03
2010 −0.55 1.13 1.13 −0.56 1.22 1.22
2020 −0.15 1.68 1.68 −0.08 2.01 2.01
2030 0.13 2.07 2.07 0.46 2.75 2.75
2040 0.31 2.37 2.37 1.07 3.57 3.57
2050 0.50 2.73 2.73 1.42 4.11 4.11
2060 0.55 2.89 2.89 2.08 4.92 4.92
2070 0.41 2.95 2.95 2.74 6.18 6.18
2080 0.30 2.83 2.83 3.24 6.73 6.73
2090 0.38 2.94 2.94 3.93 7.61 7.61
2100 0.44 3.02 3.02 4.33 8.16 8.16
EPU_SS - Surface Temperature
2000 −1.09 0.86 0.86 −1.09 0.86 0.86
2010 −0.68 1.03 1.03 −0.69 1.09 1.09
2020 −0.24 1.59 1.59 −0.17 1.88 1.88
2030 0.02 1.92 1.92 0.37 2.58 2.58
2040 0.26 2.23 2.23 1.00 3.34 3.34
2050 0.44 2.64 2.64 1.35 3.90 3.90
2060 0.51 2.74 2.74 2.00 4.61 4.61
2070 0.33 2.76 2.76 2.70 5.84 5.84
2080 0.21 2.71 2.71 3.18 6.35 6.35
2090 0.26 2.78 2.78 3.86 7.21 7.21
2100 0.34 2.69 2.69 4.26 7.67 7.67
EPU_GOM - Surface Temperature
2000 −1.05 0.63 0.63 −1.05 0.63 0.63
2010 −0.70 0.85 0.85 −0.66 0.94 0.94
2020 −0.22 1.40 1.40 −0.14 1.71 1.71
2030 0.03 1.74 1.74 0.37 2.39 2.39
2040 0.22 2.07 2.07 1.00 3.17 3.17
2050 0.42 2.41 2.41 1.43 3.81 3.81
2060 0.49 2.48 2.48 2.08 4.67 4.67
2070 0.38 2.61 2.61 2.80 5.90 5.90
2080 0.24 2.54 2.54 3.32 6.46 6.46
2090 0.33 2.65 2.65 4.04 7.38 7.38
2100 0.43 2.44 2.44 4.44 8.05 8.05
EPU_GB - Surface Temperature
2000 −0.76 1.04 1.04 −0.76 1.04 1.04
2010 −0.41 1.23 1.23 −0.38 1.33 1.33
2020 −0.11 1.69 1.69 0.13 2.01 2.01
2030 0.15 2.05 2.05 0.60 2.67 2.67
2040 0.38 2.32 2.32 1.14 3.43 3.43
2050 0.50 2.67 2.67 1.44 3.88 3.88
2060 0.51 2.74 2.74 2.02 4.60 4.60
2070 0.37 2.67 2.67 2.69 5.74 5.74
2080 0.26 2.58 2.58 3.11 6.21 6.21
2090 0.21 2.55 2.55 3.75 7.07 7.07
2100 0.31 2.52 2.52 4.07 7.61 7.61
EPU_MAB - Surface Temperature
2000 −0.66 1.13 1.13 −0.66 1.13 1.13
2010 −0.39 1.24 1.24 −0.33 1.45 1.45
2020 −0.13 1.62 1.62 0.15 2.07 2.07
2030 0.20 1.96 1.96 0.56 2.63 2.63
2040 0.35 2.15 2.15 1.01 3.27 3.27
2050 0.42 2.48 2.48 1.40 3.73 3.73
2060 0.43 2.44 2.44 1.98 4.36 4.36
2070 0.34 2.44 2.44 2.60 5.22 5.22
2080 0.27 2.35 2.35 3.00 5.65 5.65
2090 0.25 2.27 2.27 3.61 6.40 6.40
2100 0.48 2.31 2.31 3.77 6.55 6.55
Ensemble Differences from Modern Conditions
Difference in Bias-Corrected Ensemble Data when Compared to 2010-2019 Averages from OISSTv2 & SODA
SSP1: 2.6 SSP5: 8.5
Ensemble Mean 5th Percentile 95th Percentile Ensemble Mean 5th Percentile 95th Percentile
Canadian Survey Area - Bottom Temperature
2000 −0.68 0.69 0.69 −0.68 0.69 0.69
2010 −0.27 0.96 0.96 −0.41 0.88 0.88
2020 0.17 1.68 1.68 0.08 1.77 1.77
2030 0.60 2.16 2.16 0.57 2.32 2.32
2040 0.97 2.65 2.65 1.23 3.22 3.22
2050 1.35 3.40 3.40 1.66 3.86 3.86
2060 1.56 3.69 3.69 2.37 4.69 4.69
2070 1.57 3.94 3.94 3.02 5.56 5.56
2080 1.57 4.02 4.02 3.55 6.15 6.15
2090 1.57 4.04 4.04 4.29 7.14 7.14
2100 1.69 4.17 4.17 4.66 7.76 7.76
EPU_SS - Bottom Temperature
2000 −0.73 0.44 0.44 −0.73 0.44 0.44
2010 −0.42 0.67 0.67 −0.52 0.65 0.65
2020 −0.03 1.26 1.26 −0.06 1.40 1.40
2030 0.41 1.82 1.82 0.36 1.90 1.90
2040 0.65 2.11 2.11 1.01 2.66 2.66
2050 1.00 2.92 2.92 1.43 3.29 3.29
2060 1.13 2.97 2.97 2.10 3.96 3.96
2070 1.18 3.18 3.18 2.73 5.05 5.05
2080 1.11 3.26 3.26 3.43 5.76 5.76
2090 1.13 3.33 3.33 4.22 6.82 6.82
2100 1.21 3.22 3.22 4.69 7.43 7.43
EPU_GOM - Bottom Temperature
2000 −0.76 0.50 0.50 −0.76 0.50 0.50
2010 −0.44 0.72 0.72 −0.56 0.79 0.79
2020 −0.02 1.38 1.38 −0.18 1.52 1.52
2030 0.30 1.74 1.74 0.24 2.00 2.00
2040 0.47 2.00 2.00 0.87 2.74 2.74
2050 0.84 2.78 2.78 1.29 3.34 3.34
2060 0.93 2.80 2.80 1.92 4.08 4.08
2070 0.98 3.04 3.04 2.60 5.27 5.27
2080 0.91 3.03 3.03 3.31 5.97 5.97
2090 0.93 3.02 3.02 4.07 6.73 6.73
2100 1.05 2.89 2.89 4.55 7.33 7.33
EPU_GB - Bottom Temperature
2000 −0.49 0.80 0.80 −0.49 0.80 0.80
2010 −0.16 1.14 1.14 −0.31 1.09 1.09
2020 0.22 1.80 1.80 0.10 1.80 1.80
2030 0.57 2.16 2.16 0.56 2.39 2.39
2040 0.88 2.63 2.63 1.12 3.22 3.22
2050 1.16 3.20 3.20 1.52 3.73 3.73
2060 1.35 3.42 3.42 2.13 4.49 4.49
2070 1.37 3.61 3.61 2.78 5.53 5.53
2080 1.30 3.66 3.66 3.33 6.13 6.13
2090 1.32 3.73 3.73 4.09 7.13 7.13
2100 1.40 3.86 3.86 4.33 7.63 7.63
EPU_MAB - Bottom Temperature
2000 0.04 1.67 1.67 0.04 1.67 1.67
2010 0.39 2.09 2.09 0.28 2.06 2.06
2020 0.78 2.83 2.83 0.72 2.75 2.75
2030 1.22 3.24 3.24 1.20 3.32 3.32
2040 1.43 3.67 3.67 1.65 4.05 4.05
2050 1.68 4.19 4.19 2.12 4.56 4.56
2060 1.76 4.36 4.36 2.75 5.33 5.33
2070 1.80 4.50 4.50 3.42 6.25 6.25
2080 1.71 4.51 4.51 3.87 6.81 6.81
2090 1.78 4.56 4.56 4.57 7.75 7.75
2100 1.88 4.69 4.69 4.77 8.19 8.19
Ensemble Differences from Modern Conditions
Difference in Bias-Corrected Ensemble Data when Compared to 2010-2019 Averages from OISSTv2 & SODA
SSP1: 2.6 SSP5: 8.5
Ensemble Mean 5th Percentile 95th Percentile Ensemble Mean 5th Percentile 95th Percentile
Canadian Survey Area - Surface Salinity
2000 0.05 0.89 0.89 0.05 0.89 0.89
2010 0.03 0.78 0.78 0.03 0.82 0.82
2020 −0.01 0.79 0.79 0.01 0.85 0.85
2030 −0.06 0.73 0.73 −0.06 0.84 0.84
2040 −0.13 0.82 0.82 −0.10 0.90 0.90
2050 −0.21 0.81 0.81 −0.32 0.68 0.68
2060 −0.33 0.67 0.67 −0.36 0.70 0.70
2070 −0.37 0.65 0.65 −0.49 0.74 0.74
2080 −0.39 0.68 0.68 −0.71 0.62 0.62
2090 −0.47 0.66 0.66 −0.88 0.55 0.55
2100 −0.37 0.95 0.95 −0.97 0.57 0.57
EPU_SS - Surface Salinity
2000 0.03 0.96 0.96 0.03 0.96 0.96
2010 0.01 0.89 0.89 0.03 0.92 0.92
2020 0.03 0.94 0.94 0.06 1.06 1.06
2030 −0.02 0.87 0.87 −0.01 1.05 1.05
2040 −0.04 1.03 1.03 −0.01 1.13 1.13
2050 −0.11 1.06 1.06 −0.25 0.86 0.86
2060 −0.24 0.89 0.89 −0.24 0.89 0.89
2070 −0.28 0.88 0.88 −0.34 1.05 1.05
2080 −0.29 0.85 0.85 −0.58 0.90 0.90
2090 −0.39 0.92 0.92 −0.74 0.81 0.81
2100 −0.30 1.23 1.23 −0.84 0.89 0.89
EPU_GOM - Surface Salinity
2000 −0.02 0.85 0.85 −0.02 0.85 0.85
2010 −0.04 0.78 0.78 −0.04 0.80 0.80
2020 −0.02 0.81 0.81 −0.02 0.89 0.89
2030 −0.09 0.76 0.76 −0.08 0.89 0.89
2040 −0.11 0.92 0.92 −0.08 0.99 0.99
2050 −0.17 0.92 0.92 −0.28 0.69 0.69
2060 −0.30 0.78 0.78 −0.26 0.75 0.75
2070 −0.31 0.80 0.80 −0.35 0.89 0.89
2080 −0.33 0.71 0.71 −0.58 0.75 0.75
2090 −0.41 0.80 0.80 −0.73 0.68 0.68
2100 −0.31 1.12 1.12 −0.79 0.77 0.77
EPU_GB - Surface Salinity
2000 −0.17 0.64 0.64 −0.17 0.64 0.64
2010 −0.11 0.63 0.63 −0.11 0.69 0.69
2020 −0.11 0.73 0.73 0.00 0.88 0.88
2030 −0.10 0.75 0.75 −0.03 0.93 0.93
2040 −0.10 0.88 0.88 −0.01 1.05 1.05
2050 −0.19 0.90 0.90 −0.23 0.76 0.76
2060 −0.30 0.76 0.76 −0.20 0.82 0.82
2070 −0.32 0.71 0.71 −0.25 0.95 0.95
2080 −0.35 0.69 0.69 −0.44 0.84 0.84
2090 −0.50 0.69 0.69 −0.53 0.84 0.84
2100 −0.42 0.96 0.96 −0.58 0.89 0.89
EPU_MAB - Surface Salinity
2000 −0.20 0.44 0.44 −0.20 0.44 0.44
2010 −0.15 0.42 0.42 −0.14 0.48 0.48
2020 −0.14 0.47 0.47 0.01 0.72 0.72
2030 −0.06 0.58 0.58 0.02 0.74 0.74
2040 −0.02 0.71 0.71 0.05 0.91 0.91
2050 −0.08 0.72 0.72 0.00 0.79 0.79
2060 −0.10 0.69 0.69 0.08 0.93 0.93
2070 −0.11 0.70 0.70 0.14 1.09 1.09
2080 −0.13 0.69 0.69 0.08 1.09 1.09
2090 −0.22 0.73 0.73 0.06 1.17 1.17
2100 −0.14 0.85 0.85 0.04 1.38 1.38
Ensemble Differences from Modern Conditions
Difference in Bias-Corrected Ensemble Data when Compared to 2010-2019 Averages from OISSTv2 & SODA
SSP1: 2.6 SSP5: 8.5
Ensemble Mean 5th Percentile 95th Percentile Ensemble Mean 5th Percentile 95th Percentile
Canadian Survey Area - Bottom Salinity
2000 −0.07 0.28 0.28 −0.07 0.28 0.28
2010 −0.04 0.26 0.26 −0.04 0.27 0.27
2020 −0.03 0.31 0.31 0.01 0.39 0.39
2030 0.03 0.38 0.38 0.06 0.44 0.44
2040 0.08 0.48 0.48 0.12 0.56 0.56
2050 0.11 0.56 0.56 0.15 0.64 0.64
2060 0.10 0.57 0.57 0.25 0.74 0.74
2070 0.12 0.64 0.64 0.30 0.86 0.86
2080 0.13 0.65 0.65 0.32 0.92 0.92
2090 0.11 0.67 0.67 0.36 1.04 1.04
2100 0.15 0.74 0.74 0.42 1.22 1.22
EPU_SS - Bottom Salinity
2000 −0.07 0.31 0.31 −0.07 0.31 0.31
2010 −0.07 0.24 0.24 −0.06 0.26 0.26
2020 −0.06 0.29 0.29 0.01 0.39 0.39
2030 −0.01 0.36 0.36 0.03 0.40 0.40
2040 0.05 0.47 0.47 0.11 0.55 0.55
2050 0.07 0.54 0.54 0.12 0.61 0.61
2060 0.03 0.49 0.49 0.20 0.67 0.67
2070 0.04 0.52 0.52 0.25 0.79 0.79
2080 0.03 0.53 0.53 0.29 0.86 0.86
2090 0.02 0.58 0.58 0.32 0.94 0.94
2100 0.03 0.62 0.62 0.35 1.11 1.11
EPU_GOM - Bottom Salinity
2000 −0.04 0.44 0.44 −0.04 0.44 0.44
2010 −0.04 0.33 0.33 −0.03 0.35 0.35
2020 −0.03 0.38 0.38 0.01 0.50 0.50
2030 0.00 0.45 0.45 0.03 0.51 0.51
2040 0.04 0.61 0.61 0.10 0.66 0.66
2050 0.04 0.66 0.66 0.07 0.66 0.66
2060 −0.01 0.55 0.55 0.14 0.70 0.70
2070 −0.01 0.56 0.56 0.18 0.83 0.83
2080 −0.03 0.53 0.53 0.17 0.85 0.85
2090 −0.05 0.62 0.62 0.17 0.85 0.85
2100 −0.07 0.66 0.66 0.18 1.02 1.02
EPU_GB - Bottom Salinity
2000 −0.08 0.35 0.35 −0.08 0.35 0.35
2010 −0.06 0.28 0.28 −0.05 0.31 0.31
2020 −0.02 0.39 0.39 0.02 0.46 0.46
2030 0.02 0.42 0.42 0.06 0.50 0.50
2040 0.10 0.59 0.59 0.13 0.67 0.67
2050 0.09 0.63 0.63 0.16 0.69 0.69
2060 0.08 0.61 0.61 0.24 0.74 0.74
2070 0.08 0.64 0.64 0.29 0.90 0.90
2080 0.09 0.66 0.66 0.31 0.96 0.96
2090 0.07 0.71 0.71 0.34 1.10 1.10
2100 0.13 0.82 0.82 0.38 1.25 1.25
EPU_MAB - Bottom Salinity
2000 −0.06 0.40 0.40 −0.06 0.40 0.40
2010 −0.01 0.38 0.38 0.00 0.41 0.41
2020 0.01 0.44 0.44 0.08 0.55 0.55
2030 0.09 0.52 0.52 0.14 0.61 0.61
2040 0.17 0.69 0.69 0.18 0.78 0.78
2050 0.17 0.74 0.74 0.26 0.83 0.83
2060 0.16 0.76 0.76 0.36 0.95 0.95
2070 0.17 0.81 0.81 0.45 1.13 1.13
2080 0.18 0.85 0.85 0.49 1.26 1.26
2090 0.18 0.92 0.92 0.55 1.43 1.43
2100 0.22 1.02 1.02 0.57 1.62 1.62

Distance from Modern Conditions Timelines

For this section I wanted to play with the intra-annual variation. Are springs further from modern conditions etc. ? I also wanted to showcase cases where the transition is not linear

Aside: Exporting Rasters as Netcdf